Anthropic released Claude Opus 4.6 on February 5, 2026 — just three months after Opus 4.5. It’s a significant upgrade. Here’s what changed.
⚡ Update (April 17, 2026): Anthropic has since released Claude Opus 4.7, which jumps to 64.3% on SWE-bench Pro, adds 3.75MP vision, and introduces the xhigh effort level. If you’re choosing between models today, see our Opus 4.7 vs 4.6 comparison.
Key changes
| Opus 4.5 | Opus 4.6 | |
|---|---|---|
| Release date | November 2025 | February 5, 2026 |
| Context window | 200K tokens | 1M tokens (beta) |
| Max output | 8K tokens | 128K tokens |
| SWE-bench Verified | ~76.8% | ~80.8% |
| MRCR v2 (1M, 8-needle) | N/A (200K limit) | 76% |
| Adaptive thinking | No | Yes |
| Agent teams (Claude Code) | No | Yes |
| Input price | $5 / 1M tokens | $5 / 1M tokens |
| Output price | $25 / 1M tokens | $25 / 1M tokens |
What improved
Context window: 200K → 1M tokens. This is the headline feature. Opus 4.6 is the first Opus-class model to support 1 million tokens of context, letting you process entire codebases or books in a single prompt. Note: long context requests double the input price.
Max output: 8K → 128K tokens. Opus 4.6 can generate much longer responses — useful for generating entire files, long analyses, or detailed documentation.
Coding (SWE-bench): 76.8% → 80.8%. A solid improvement in autonomous coding ability, with better multi-file understanding and fewer logic errors.
Adaptive thinking. Opus 4.6 can dynamically adjust how much reasoning it applies based on task complexity — spending more time on hard problems and less on simple ones.
Collaborative agent teams. In Claude Code, Opus 4.6 can coordinate multiple agents working on different parts of a codebase simultaneously.
What stayed the same
- Pricing ($5/$25 per million tokens)
- Multimodal support (text + images)
- API compatibility (drop-in replacement)
Should you upgrade?
Absolutely. Same price, bigger context, better coding, new features. There’s no reason to stay on 4.5. Just swap the model ID in your API calls.
FAQ
Is Claude Opus 4.6 worth upgrading from 4.5?
Absolutely. Same price, bigger context (1M vs 200K), better coding (80.8% vs 76.8% SWE-bench), 128K max output (vs 8K), adaptive thinking, and collaborative agent teams. There’s no reason to stay on 4.5 — it’s a strict upgrade at the same cost.
What changed in Opus 4.6?
The biggest changes are the 5x context window increase (200K → 1M tokens), 16x max output increase (8K → 128K tokens), a 4-point SWE-bench improvement, adaptive thinking that adjusts reasoning depth per task, and collaborative agent teams in Claude Code.
Is Opus 4.5 still available?
Opus 4.5 is still accessible via the API using its model ID, but there’s no practical reason to use it. Opus 4.6 is a drop-in replacement at the same price with better performance across every metric. Note that Anthropic has since released Opus 4.7 with even further improvements.
See our full AI Model Comparison for all models side by side.